[patch 0/8] systemsim and hvc_console patches

[Eric Van Hensbergen]

On 1/15/06, Arnd Bergmann <arndb at de.ibm.com> wrote:
> > Overall I'm not sure how I feel about having hacks in the kernel for
> > systemsim. If it's truly simulating the full system, why can't it simulate
> > the disk device and network device as well instead of doing things with
> > magic instructions like it does now?
>
> I think systemsim does have accurate models for a number of IDE and
> network devices, but I would rather not have to use them because they will
> run at a horribly slow speed, because they try to model the performance
> of the devices relative to the CPU cycles.
>

We could probably produce some performance numbers on this from the
simulator side, but the overall reports from the end-users were that
non-bogus devices were really slow.  That being said, I see no reason
why they can't be available as a patch -- most folks using the
simulator are building their own kernels anyways.

>
> For the cell simulator in particular, I think there is no model of the
> spider-pic interrupt controller, which means that systemsim can not use
> the models for the devices it uses on other platforms. Eric, is this still
> correct?
>

I believe we have a model for the interrupt controller, but I'm not
sure what its releasability status is from the STI-end of things. 
IIRC the real complication when we tried to use it (even for
delivering interrupts to BogusNet) was problems with the firmware not
being set up correctly to route the interrupts.   This results in the
Cell BogusNet having to run in polled mode (and having pretty poor
performance as a result).  There were some attempts by the simulator
team to resolve this issue, but I'm not sure where they left it (so
I'm copying them on this message to see if we can get a definitive
answer).

          -eric